A Practical Guide to Claude Code for Backend Engineers

Claude Code has become my default for backend work that involves more than a single file. Here is the workflow I settled on after a few false starts.

The setup that matters

Two things made a real difference for me:

• A clean AGENTS.md that lists the project conventions, the test command, and the things it must never touch.
• A Makefile target that runs the full check, not just the unit tests, so the model sees failures it can react to.

The shape of a good task

I try to write every prompt as a small story: what I want, why I want it, and how I will know it works. Something like:

Add a POST /v1/sessions/:id/cancel endpoint. It should be idempotent, return 204 on success and 404 if the session does not exist. Add tests for both paths and make sure the existing webhook tests still pass.

The more concrete the success criteria, the less I have to argue with it later.

When I do not use it

Anything that touches money, auth tokens, or production data does not get pasted in. The risk is not worth the time saved.

Happy to compare notes if you have a different setup that works for your team.